Research On Sparse Sampling And Outlier Detection Mechanism Of Agricultural Data

The agricultural Internet of Things system has become one of the most important sources for agricultural big data.By means of deploying micro-sensors with sensing,communication,and computing capabilities in agricultural production management,it is achieved accurately,and efficiently to monitor the Soil-Plant-Atmosphere Continuum system for helping the realization of precision agriculture.However,due to the limitation factors such as working environment,equipment manufacturing,and network link conditions,it is challenging to implement the deployment and application of agricultural Internet of Things and high-quality data acquisition.Therefore,it is still urgent to develop the method of low-power,high-quality agricultural data collection.Aiming at reducing low power consumption of agricultural sensors and offering real-time perception guarantee of key data and outlier detection,this study focuses on several following issues,such as,node sparse sampling decision,energy consumption optimization,important data perception guarantee,outlier detection.This research may provide a reference for reducing energy consumption and improving the quality of collected agricultural data.The main research content and results are as follows,1)Research on sparse sampling decision methods of agricultural monitoring nodes based on compressed sensing.Compressed sensing is applied to the sparse sampling of agricultural condition monitoring nodes,and a node sparse sampling assisted decision-making system based on compressed sensing is developed.The measurement matrix,sparse basis,reconstruction algorithm,and data reconstruction interval suitable for agricultural condition sampling were systematically analyzed.Through applications of the developed system,several following conclusions are drawn.When conducting compressive sensing for 30-day continuous monitoring of plant sap flow and soil moisture with a15-minute sampling interval,it is appropriate to choose periodical measurement matrix,difference matrix,and SL₀ as measurement matrix,representation basis,and reconstruction algorithm.Additionally,one useful compromise between data recovery cost and accuracy could be achieved if reconstructing data every 2-10 days.2)Data collection mechanism for energy consumption optimization and a critical-sensing guarantee.Combining compressed sensing and model driving,an adaptive switching mechanism of data acquisition mode is proposed.By merging key data thresholds and prediction error tolerance,the moment of switching from model-driven mode to compressed sensing mode is determined,and the moment of switching from compressed-sensing mode to model driving mode is predicted based on sparse sampling point modeling.Experimental results show that the proposed method can effectively avoid the jitter of data collection mode switching,ensure the accuracy of data acquisition,and the timeliness of key data acquisition.3)Real-time outlier detection method of agricultural data by means of increasing vector dimension.The normalized time-series agricultural data is converted into observation vectors,and then the dimension increase of observation vector is implemented by inversely summing the difference of vector elements,and a real-time detection framework of agricultural data based on vector dimension increase is constructed.The results of outlier detection in conventional and sparse sampling scenarios show that the proposed method can effectively improve the outlier detection effect,and support vector machine with linear kernel outperforms the counterparts in terms of outlier detection performance and computation overhead.